Why are we creating the matrix 'weight' of 2 dimensions i.e 2*2 when there are only 2 features, like we did for the biases?